knowledge: improve review precision from BCApps PR 11063 feedback

[Wenjie Fan (gggdttt)]

Summary

Improves BCQuality knowledge based on maintainer thumbs-down feedback from explicit BCApps PR review runs.

Source feedback

• Prevent Sales Order Agent from skipping emails moved into monitored folders BCApps#11063 (comment)

Validation

• Frontmatter and article structure validation completed.

Generated by the BC-ALAgentsInternal self-improvement workflow.

Offline evaluation: regression

Candidate correctness failed: unexpected or missed findings remain. Existing ignored gold comments retain their neutral scoring semantics.

Common engine code: ecf8e31759d6ddd6d78e3a0b7836b40134368009; dataset SHA256: ACDC50C36CC88DD31A5AB628A7C3D06A44E5F7628DA9DA79A0916E6AF0F46222.
Model: gpt-5.6-luna; judge: gpt-5.3-codex. Exact D: synthetic__perf-watermark-folder-sync-01.
Dataset base: 231538937b19317eddc3ac654459ca29b3ef4170; candidate: 46b6ad94497fcb272417b170b2cbb7b3f96d7209.

Arm	Evaluation commit	Engine pin	Knowledge pin
baseline	27da12f4d5dfec9862f23219b9d1e80c465cc18a	4407f20b44fe92e3693d0444d25cc3430f1d5262	8584217c7506eea7eef27a9db353d78c0cd6a9a1
candidate	09695a2fecf379e823762ccbc425a09b9259a823	35b7bcdecf431b9b1a18ca83830a38c2fc353344	e74e2a6601ec87a1ffec6e3d7eb074aa67437ef2

Baseline

Run: https://github.com/microsoft/BC-Bench/actions/runs/34221111997; conclusion: success; wall clock: 12.6 minutes.
Raw findings and runtime pin proofs are retained in the self-improvement-feedback artifact and the linked evaluation run.

Entry	Expected	Generated	Missed	Unexpected	F1	AI credits	Agent seconds
synthetic__perf-watermark-folder-sync-01	0	2	0	2	0	unavailable	572.746033486

Observed AI credits: unavailable; coverage 0/1 entries. Missing entries are not extrapolated into a total.

Aggregate metric	Value
total	1
expected_comment_count	0
generated_comment_count	2
matched_comment_count	0
missed_comment_count	0
incorrect_comment_count	2
precision	0
recall	1
f1	0

Candidate

Run: https://github.com/microsoft/BC-Bench/actions/runs/34222250479; conclusion: success; wall clock: 7 minutes.
Raw findings and runtime pin proofs are retained in the self-improvement-feedback artifact and the linked evaluation run.

Entry	Expected	Generated	Missed	Unexpected	F1	AI credits	Agent seconds
synthetic__perf-watermark-folder-sync-01	0	1	0	1	0	unavailable	234.327608426

Observed AI credits: unavailable; coverage 0/1 entries. Missing entries are not extrapolated into a total.

Aggregate metric	Value
total	1
expected_comment_count	0
generated_comment_count	1
matched_comment_count	0
missed_comment_count	0
incorrect_comment_count	1
precision	0
recall	1
f1	0

Missing telemetry is unavailable, not zero. Evaluation-only metrics exclude candidate generation and are not the full-cycle cost.
Human review must verify source-patch fidelity, gold correctness, and target attribution. No automatic merge or branch-protection claim is made.

[Jesper Schulz-Wedde (JesperSchulz)]

Two changes are necessary before this precision rule can land:

1. The supplied exact regression still fails: the candidate generates 1 unexpected finding for synthetic__perf-watermark-folder-sync-01 (expected 0, F1 0). It only reduces the original two false positives to one. Please resolve the remaining deterministic route/evaluator result, or correct the fixture/gold if that remaining finding is independently valid, then rerun to a clean result.
2. This stale branch no longer passes the current exact-tree contract. Merging it with current main (852a676; computed merge tree bd8b8f4) makes tools/Test-ReviewFixtures.ps1 fail: both companion samples are code spans rather than READ-convention links. Use [filename.good.al](filename.good.al) and the equivalent .bad.al link.

I traced BCApps #11063's thumbs-down and replies: the conditional rule is otherwise correct—the received-time watermark is unsafe when older items can become newly visible, while a server-side processed-state filter prevents those items consuming $top; monotonic change/delta tokens remain explicitly excluded. The AL examples preserve that boundary, [all] is appropriate, current checks and exact-head frontmatter/index/34-fixture validation pass, and current-main merged-tree frontmatter/index validation passes. The merged-tree fixture failure and unresolved 1/1 false positive are the blockers.